| Low altitude UAV remote sensing has insurmountable advantages over conventional aerial remote sensing system, including low cost, quick response, high image resolution and so on. Currently, it plays an irreplaceable role in disaster forecast and warning, emergency response, military reconnaissance and other fields. However, some problems can also be found in UAV remote sensing images, such as serious distortion, small frame size, different gray-level and scales and others. Consequently, appropriate algorithm should be chosen to develop the UAV image stitching system in order to obtain real-time clear images from broader angles of view.Scale Invariant Feature Transform algorithm (SIFT) is robust in rotation and zooming, scale transformation, changes of brightness and view and so on. Therefore, based on the study of basic theories and key technology of image stitching, this paper first analyzes the features of SIFT and reviews the problems in parameter setting and false matching. Then it proposes using signal & slots mechanism and function transfer to solve the problem of independent setting of parameters and threshold and makes an experimental analysis. Using neighbor domain voting method to improve the algorithm combined with robust estimation algorithm to purify the matching points, a systematic framework and process of design is given with various modules, including feature detection, matching and purification, stitching processing, and interface input and output. Finally, based on the improved SIFT algorithm, a UAV image stitching software is developed, with which interfaces can be interacted, thresholds can be adjusted in real time and results can also be saved and exported.The results show that improved SIFT algorithm can filter false matching points effectively, ensuring the image registration accuracy while improving the efficiency of matching purification. By adopting the idea of modularization to design and develop, the stitching system has strong extensibility. Since the stitching system is characteristic of friendly threshold adjustment and result output interfaces, the users may choose the appropriate threshold according to different application purposes and analyze the exported results, which shows great availability of the system. Generally speaking, image stitching software proposed in this paper can get mosaic image conveniently and quickly, and prove to be highly effective in use. |